Research On The Performance Optimization Method Of Electronic Nose Based On Gas Detection

The development of modern society has brought about an increase in productivity,which has greatly facilitated the production of human life.However,the indoor and the natural environment around us are also subject to a considerable degree of pollution,and the monitoring of Volatile Organic Compounds（VOCs）is one of the major concerns of scientists today.The electronic nose is an intelligent machine consisting of gas sensors,systems of data pre-processing and pattern recognition algorithms to simulate human olfactory functions,which can detect these substances very well.The Thesis focuses on the problem of VOC mixture detection and explores the gas sensor by combining the approaches of sensor arrays and pattern recognition techniques.The main work is as follows:（1）Sensing elements of various gas-sensitive materials are respectively prepared by photolithography and drop coating methods combined with interdigital electrodes.The Zn O gas sensors prepared by drop-coating method are tested for gas sensitivity.The experiments show that the devices obtain excellent gas response at room temperature under UV,with a detection limit of 40 ppb and a good response repeatability over a certain operating temperature range.A controlled preparation method for controlling the morphology of the gas-sensitive functional layer is also proposed based on chemical vapor deposition.A series of characterizations show that the thermal aggregation of Ti O₂films as the induced layer induces a gradient strain effect that can lead to the extrusion and upward growth of Mo S₂ films,resulting in the growth of Mo S2 vertical nanowalls structures,which helps to enhance the responsiveness and sensitivity of gas-sensitive elements in gas detection.（2）A gas classification model based on Back Propagation Neural Network is designed.Firstly,the feature data is pre-processed by normalization and the high-dimensional data is dimensionally reduced.The number of iterations in BPNN and various gradient optimization algorithms are investigated to determine the effect on the classification performance of BPNN,and the experimental results were compared with other machine learning algorithms which shows that the model can achieve high accuracy in classifying mixed gases and outperforms other traditional algorithms.（3）In this thesis,with the aim of improving the classification accuracy of the model,the method of pattern recognition is improved.A BPNN gas classification model based on an ensemble learning algorithm is designed.Firstly,an optimized feature reduction method is used to process the data,a random oversampling method is used to optimize the imbalance of various types of data in the dataset,the misclassified data and the weaker classifier weights are adjusted by the Ada Boost algorithm,and the final strong classifier is obtained by the combination of the weighted weaker classifiers.The experimental results show that the improved dimensionality reduction method can reduce the training time of the gas classification recognition model with the same network parameters,and the integrated learning algorithm can effectively improve the recognition accuracy of the original model,with the accuracy rate increasing by more than 2 percentage points,reaching 98.10%on average and 98.99%at the highest compared with other pattern recognition algorithms.